Performance comparison of cloud virtual machines Open Access

In the past, organizations simply purchased and maintained physical hardware if they needed to deploy a new server. This approach has evolved and the current trend in deploying a server is to employ virtualization/cloud computing technology. The use of these technologies assists in accelerating deployment time and facilitating maintenance. A vast number of virtualization and cloud service options are available to organizations, and therefore, selecting the most suitable provider is a challenging task.

Comparing the performance of the virtual machine (VM) is an area which might be scrutinized before an organization selects a provider. A performance comparison might also influence the decision whether to use cloud services or keep an on-premise server. Computational performance is a critical factor, as it determines how many operations a machine is able to process in a certain period. Depending on the particular system, suitable performance may also involve server availability, bandwidth, latency, response time and many other aspects as was described by Fuller and Millett (2011). Besides performance considerations, selecting the most suitable cloud provider for a VM involves other aspects and obstacles. As the significant ones, Shirvani (2020) selected security characteristics, costs/licences, reliability and integration options. Many researchers have already established a complex decision model for considering whether to use cloud services. Shirvani et al. (2018) published an example in which the authors described a mathematical model for cloud migration based on a cost and security risk approach. Another perspective provided Ma et al. (2023) where the authors focused on migration strategy of VMs with the usage of Markov decision and greedy algorithm.

Using VMs is not the only option for organizations to make use of virtualized computational power. The deployment of containers is a widespread solution, especially popular with developers. Its principal feature is a VM running under a hypervisor as a standalone computer with its own operating system (OS). By contrast, containers are abstracted at the OS/kernel level and multiple containers might run simultaneously on the same OS. Abstraction assists developers in moving applications seamlessly from one environment (machine, OS, libraries) to another without producing any interoperability issues as described by Celesti et al. (2019). Multiple tools also exist for the deployment of container solutions. Pratap et al. (2021) presented Docker or Kubernetes as examples of common orchestration tools. Container solutions are a hot topic in cloud application deployment, resonating strongly in the research field. Zhang (2024) focused on optimizing container orchestration technology for cloud computing environments. Similarly, Vaño et al. (2023) investigated container workloads and their orchestration at the “edge”. Naresh and Ayyappa (2024) explored edge computing’s role in complementing cloud services. They found that edge computing can reduce latency and enhance cloud application performance by processing data closer to its source. The current study, however, does not provide a comparison of container services since VMs have major roles in organizations regardless.

To evaluate performance of the VMs, the authors have established the following research questions on which they are responding within this paper:

Cloud computing is defined in the NIST 800–145 Definition of Cloud Computing document that was published by Mell and Grance (2011) as a model for enabling ubiquitous, convenient, on-demand network access to a shared pool of configurable computing resources (…) that can be rapidly provisioned and released with minimal management effort or service provider interaction. NIST further specifies five generic characteristics of all cloud services: on-demand self-service, broad network access, resource pooling, rapid elasticity and measured service. The document also contains a description of two well-known cloud models. The first model is called the deployment model and splits cloud services according to who has access to the services. This model involves four individual cloud service types:

1. public cloud – resources are operated in an environment where the public or other organizations also operate their own resources;

2. private cloud – resources are operated in an environment dedicated to a single organization where no other entity has access;

3. community cloud – resources are operated in an environment dedicated to a community (group) of organizations; and

4. hybrid cloud – any combination of the above-mentioned cloud types.

The second model described in the NIST document is the service model, which is characterized by three general service types –IaaS, Platform-as-a-Service and Software-as-a-Service. This model specifies the resources delivered by the cloud service provider (Figure 1). The model illustrated in Figure 1 depicts the shared responsibilities of the cloud service customer and cloud service provider as described by Singh and Sharma (2021). Because clients are highly dependent in many areas on the virtualization technology/servers of the cloud provider, the importance of sufficient operational characteristics in cloud services (resources) becomes apparent.

Correct placement of the VM in the cloud provider’s environment is a carefully managed task which ensures effective and secure deployment. The VM’s placement significantly affects the performance of the entire system and consequently the contracted service level agreement between the cloud provider and customer. From the provider’s point of view, correct placement is also important to the energy required to run computational resources. Placement of VMs in the cloud environment is not a minor consideration as cloud providers operate many data centers globally and organizations frequently decide on the regions where they want to run certain resources. The distance and connection (e.g. bandwidth) between those data centers then affects the performance of the entire system. The placement of VMs relates not only to selecting the data center, it also involves the logical architecture of systems, for example, placement within a cluster, behind network appliances and so on as described by Saeedi and Hosseini (2021). Multiple models are available for the placement of VMs within cloud data centers. Farzai et al. (2020) designed a mathematical multi-objective optimization model and performed suitable simulations and evaluations. The model’s primary objective was energy savings, less waste of resources and fewer data transfers between data centers.

Cloud computing has rapidly evolved, covering a wide range of applications and responding to emerging demands in scalability, cost-effectiveness, security and performance optimization. Several recent reviews have highlighted core trends in the field, suggesting that cloud research increasingly focuses on scalable architectures and advanced technologies, such as edge computing and artificial intelligence, to meet the complex needs of modern applications. This aligns with findings from studies on the importance of robust cloud infrastructure in handling expanding data storage, processing power and network bandwidth demands, addressing issues like latency and load balancing through innovative computational models (Jha and Chaturvedi, 2024; Smith et al., 2023). The need for scalable, adaptable cloud environments is essential for sectors that demand rapid processing and low-latency responses.

Economic and security aspects remain crucial as cloud adoption grows. Cost modeling frameworks and optimization strategies are essential for organizations seeking to balance performance and affordability, particularly as workloads scale (Khan et al., 2024). Security research continues to prioritize protecting cloud infrastructure from rising cyber threats, emphasizing the necessity for rigorous security measures and evolving protocols to guard data privacy and system integrity. Innovations like blockchain-based access control and federated learning for privacy preservation reflect these ongoing efforts to enhance security across distributed cloud environments (Punia et al., 2024; Peng et al., 2024). Furthermore, Ananthakrishna and Yadav (2023) investigate advancements in cloud security, stressing the importance of robust measures against increasing cyber threats. Their study underscores the evolving nature of security protocols and the need for ongoing improvements to protect cloud infrastructures.

Benchmarking methodologies are central to evaluating cloud infrastructure, particularly when comparing providers like AWS, Azure and GCP, which dominate the IaaS market. Performance assessments using tools such as the Phoronix Test Suite are frequently employed to measure key metrics, including CPU, memory, disk I/O and network performance. Studies using this suite have demonstrated its effectiveness in providing controlled, repeatable comparisons of computational resources across cloud providers, reinforcing its role as a reliable benchmark for cloud performance analysis (Putri et al., 2020; Milichovský, 2012; Malinowski, 2021; Páleník, 2016; Baruwal et al., 2016). Such evaluations help organizations determine the most suitable cloud solutions based on application requirements, with VMs being especially critical for compute-intensive tasks and applications that demand stable, isolated environments.

Virtualization technologies, including both VMs and containers, have distinct roles in cloud computing, each offering advantages depending on workload needs. Containers, which are particularly valued for their portability and efficiency in development environments, are abstracted at the OS level, allowing multiple containers to share the same kernel. However, VMs remain indispensable for applications that require high degrees of isolation and resource predictability. Comparative studies have analyzed these technologies in contexts ranging from workload migration to interference impacts in multi-tenant IaaS environments, highlighting the different scenarios in which each excels (Tay et al., 2017; Prashant, 2016; Derdus et al., 2019).

VMs are offered by all of the major cloud services providers who operate the IaaS model. In terms of providing an environment where organizations can opt to pay for computational power, the biggest cloud service providers are currently Microsoft with the Microsoft Azure platform, Amazon with the AWS platform and Google with GCP. Through their analyses, Statista (2021), Gartner (2021) and Dignan (2021) also reached this same conclusion. Microsoft, Amazon and Google each offer practically identical cloud services such as VMs, databases, storage, networking, serverless services and others.

In another comparative study, Dutta and Dutta (2019) compared the Microsoft, Amazon and Google platforms. The research mainly consisted of a comparison of services offered in areas such as databases, management and networking. The authors also identified the main benefits and drawbacks of each platform but did not indicate which platform was the most suitable. Bakaraniya and Patel (2021) explored an equivalent topic and approach.

Many other researchers have also compared these three platforms, but mainly at the theoretical level and in connection with the services and options offered by each provider. Saraswat and Tripathi (2020) compared the cloud services offered by Microsoft, Amazon and Google according to a variety of criteria, such as regional availability, range of services, security and price model. The authors concluded that AWS is the best option if organizations require a platform with the broadest, reliable and stable services and pricing is not a major challenge. The authors also concluded that Azure is the most suitable platform for its many integrated services if an organization operates Windows machines. The GCP platform is ideal if an organization is a start-up and medium-sized company and rapidly scaling up with a large volume of user data, innovation and resources.

Srinivas and Badrinarayana (2017) compared Microsoft Azure with AWS; however, the paper provides more a description of the platforms and example services than a comparison. The authors mention only high-level scenarios in which each platform appears to be a more suitable option. Scenarios in favor of selecting AWS include robust applications and the most reliable feature set. The best scenarios in relation to Microsoft Azure are integration with other Microsoft products and zero maintenance solutions. Gandhi and Kumbharana (2018) published similar research that also examined platforms only provided by Microsoft and Amazon. In contrast to Srinivas and Badrinarayana (2017), the authors mainly compared the services offered and provided little description; the authors did not offer conclusions on the more suitable platform but highlighted the advantages of each.

Srivastava et al. (2018) explored IaaS in a comparative study which applied a hybrid intuitionistic fuzzy technique in combination with TOPSIS, which is a multi-criteria decision analysis method for specialized weighted comparison of multiple alternatives. The authors compared Amazon, Microsoft, Google, DigitalOcean and IBM. Some of the criteria assessed were reliability, security, cost and scalability. The authors concluded that Amazon, Google, IBM, Azure and DigitalOcean were the most suitable platforms.

If an organization does not want to operate its own virtualization hypervisor, they have the option to purchase computing power from a cloud provider. The provider will offer a similar set of OS versions, although it might optimize (tune) the operation of machines in a different manner.

The review of published literature in the previous section did not list any comprehensive study that compares the performance of cloud VMs. The biggest providers of cloud VMs (IaaS) are Microsoft (Azure), Amazon (AWS) and Google (GCP). The products offered by those companies are generally perceived as services with the same level of quality. Our research question in the current study, therefore, is whether the VMs offered by these three cloud providers differ, and if so, how significantly.

As, to the best of our knowledge, no comprehensive study has been performed in this area, the current study’s aim is to compare the VM performance (IaaS) offered by the biggest public cloud providers: Microsoft (Azure), Amazon (AWS) and Google (GCP). We examined performance from multiple perspectives, such as CPU, memory, disk read/write operations, and so on, which are further described in Section 2.

This section describes the VM selection process, the performance comparison tool and the comparison methodology.

Although Microsoft is the direct provider of Windows OS (Windows Server 2019, Windows 10, etc.), the performance of this OS type might be artificially tuned to Microsoft Azure rather than AWS or GCP, therefore Linux was selected as the reference OS. Ubuntu 20.04 TLS was selected as the reference Linux distribution because it is offered by all three providers and is one of the most widely used. The components of the VMs, including their configurations, were set to match as closely as possible. Table 1 shows the characteristics of the selected VMs. All the VMs were deployed for the purposes of testing only and no other activity was performed on those machines.

The VMs were tested using Phoronix Test Suite, an open-source testing tool under the GNU GPLv3 licence (free for professional or personal purposes):

The tool offers hundreds of tests. The tests selected for the performance comparison in the current study are shown below. Many of the tests involved multiple specific subtests; for example, blogbench performs two subtests (one for reading data and one for writing data). Some of the tests involved only a single subtest (e.g. build-linux-kernel).

These referenced tests are sufficiently representative to provide results for adequate comparison. Each of the tests examines a different aspect of VM performance, for example, processor performance, network performance, memory processing, data processing, etc. Altogether, these tests examine the major areas of VM performance and are therefore suitable for an objective comparison of different VMs.

The results of the tests are documented in the tables in Section 3. Each record contains the test name along with the unit clarification of what the measured number indicates. The subsequent columns then contain values for each of the cloud platforms (Azure, AWS, GCP). In the majority of tests, higher figures indicate a better result, but in some tests, lower figures indicate the better result; the relevant rule is specified in the final column of each table, where “M” refers to the first, more frequent condition, and “F” refers to the latter.

For each subtest, the best and worst results are respectively highlighted in green and red. The tables also indicate comparative information (in percentages) of the differences in results between the cloud platforms. This information is indicated in the columns designated by the “%” sign. The average scores of each subtest (in percentage) were calculated from the following formula:

This calculation is valid only for tests with the “M” rule. The average scores for tests where lower figures are better were calculated from the formula:

An overall comparison is then obtained from the weighted average of the subtests. Certain tests consist of many specific subtests (e.g. fio with 32, cryptsetup with 14, ramspeed with 10) compared to others with only a few subtests (e.g. build-kernel-linux with 1, gzip with 1, himeno with 1). The tests with many results would therefore artificially impact the final comparison score. The first step was to calculate an average score for each test (fio, ramspeed, 7zip etc.), as follows:

The final score for each platform was then calculated as an average of the specific averages of the individual tests:

This section compares the results, which were divided according to the specific testing areas and their designation in Phoronix Test Suite categories. The results are summarized in the end of this section.

The processor (CPU) provides the computational power of the machine and interprets the majority of executed commands. Computation involves, for example, logical, arithmetic or I/O operations.

The tests 7zip and gzip examined the compression capability of the VM’s CPU and handling of large files during subtests. 7zip also tested the decompression of files (Phoronix Media). In 7zip the performance values are measured in MIPS, referring to millions of instructions per second; gzip measures performance in seconds. The build-linux-kernel test measures the time (in seconds) required to build the Linux kernel instance in a default configuration (Acher et al., 2019). The final test, himeno, applies an iterative algorithm called the point-Jacobi method to analyze performance by solving Poisson’s equations (HPC Advisory Council, 2010).

The measurements listed in Table 3 indicate that the AWS VM achieved the best results in all six of the tests, with an average score of 100%. The average score for Azure was 84%, and for GCP, 79%. The average scores were calculated according to Formulas 1 and 2.

The memory component temporarily stores data and instructions for immediate use or execution by other machine components. Two tests which examined memory performance were run. The ramspeed test measures the performance of the operational memory and values in MB/s. This test performs a number of subtests, as shown in Table 4. The second t-test1 benchmarks the memory allocator, measuring values in seconds (Phoronix Media).

The results in Table 4 indicate that the AWS VM achieved the best results, with an average score of 94%. In 2 t-test1 tests, however, the Azure VM outperformed the AWS VM (average score for the Azure VM was 89%). The GCP VM produced the worst results in all subtests, one with an average score of 75%.

The OS component manages all the resources (hardware and software) present in the machine. The most well-known OS are from the Windows or Linux families. The sockperf test examines both the OS and network (categorized under “OS” in Phoronix Test Suite) and measures system latency and throughput. The perf-bench test analyses Linux systems and performs seven subtests, shown below in Table 5.

This test produced varied results and is the only test where each platform was the best in terms of performance in at least one subtest. The GCP VM produced the best results at 89%; the AWS VM yielded an average score of 76%; the Azure VM achieved 70%.

A hard disk drive is a natural component of every VM. In the current study, each VM used an SSD. The initial test blogbench operated with a large file server and measured multiple-thread random reads and writes (Phoronix Media, 2024a, 2024b). The second test fio measured disk performance in relation to specific data workloads. The results are shown in Table 6 (Read The Docs, 2024).

The AWS VM produced the best results in all disk tests. This category is also one of the two categories where only one platform was best in all subtests. The remaining two platforms achieved the following scores: Azure VM (40%), GCP VM (39%).

Phoronix Test Suite has many tests which examine the overall system (e.g. software) but not components directly linked to the OS. These tests often measure other Phoronix Test Suite categories, for example sysbench, which has two subtests that examine memory and the processor. The final test was cryptsetup, which measured the cryptography performance of the VM.

The GCP VM produced the best overall system results in Phoronix Test Suite, with an average score of 99%. The Azure VM produced the worst results at 74%. The average score for the AWS VM was 79%. The results for the system (and previous) category were significantly affected by both tests (sysbnech, cryptsetup) containing a large difference in the number of specific subtests (Table 7).

The network is considered secondary for comparative purposes because it also depends greatly on many factors other than the components, configuration and tuning of the VM. Only the network test network-loopback was performed, measuring TCP performance (Table 8).

The GCP VM achieved the best results in the network-loopback test, with 100%. The AWS VM achieved 76%; the Azure VM achieved 71%.

The previous sections presented the results for specific Phoronix Test Suite categories. The following section presents the overall results and final average scores for each platform. The results were calculated according to Formula 4 and are shown in Table 9.

This section provides readers with author’s discussion and description of theoretical and practical implications of this research.

This study offers valuable insights into VM performance across three major cloud service providers: AWS, Microsoft Azure and GCP. Results show that AWS consistently outperforms Azure and GCP in key areas such as CPU processing, memory handling and disk I/O. These findings align with Saraswat and Tripathi (2020) conclusions, which identified AWS as the most reliable and stable platform for organizations needing robust cloud services. The superior performance of AWS VMs, as demonstrated here, supports the idea that AWS is well-suited for high-performance and compute-intensive applications – a finding that echoes the broader literature on cloud service performance.

The performance metrics obtained from the Phoronix Test Suite highlight the critical role of computational efficiency in cloud environments. Studies by Fuller and Millett (2011) and Shirvani (2020) have emphasized the importance of computational performance in determining the operational efficiency of cloud services. The current study’s findings reinforce these perspectives, showing that AWS’s higher performance metrics can lead to more efficient and reliable cloud operations. This is particularly relevant for organizations that rely on real-time processing and high-throughput applications, such as those in financial services, health care and e-commerce.

The integration of AI and machine learning for cloud performance optimization, as discussed by Nerella et al. (2024), also plays a crucial role in enhancing cloud efficiency. The current study’s use of benchmarking tools like the Phoronix Test Suite provides a controlled and repeatable method for assessing cloud performance, which is essential for implementing AI-driven optimization strategies. The findings suggest that integrating AI with cloud management tools can further enhance resource allocation and operational efficiency, a conclusion that aligns with Nerella et al.’s (2024) research.

The comparative analysis also reveals that while Azure and GCP offer competitive performance, they fall short of AWS in several key areas. This is consistent with the findings of Dutta et al. (2019) and Bakaraniya and Patel (2021), who noted that while Azure and GCP provide a range of integrated services and are suitable for specific use cases, they do not match AWS’s overall performance. The study by Smith et al. (2023) on scalability challenges in cloud computing further supports this, highlighting the need for scalable architectures to handle growing demand. AWS’s superior scalability, as evidenced by its performance in this study, underscores its capability to meet the needs of large-scale and dynamic workloads.

Security and cost considerations are also critical factors in cloud provider selection. The study by Shirvani (2020) on cloud adoption decision models highlights the importance of evaluating total cost of ownership (TCO) and security measures in cloud environments. While the study focuses primarily TCO, the authors significantly involves also the security aspect and underscore the necessity of a comprehensive cost-benefit analysis to ensure that cloud adoption aligns with organizational goals, financial constraints and security requirements. AWS’s strong performance, combined with its comprehensive security features, makes it a compelling choice for organizations with stringent security requirements. In addition, the cost modeling and optimization insights provided by Khan et al. (2024) suggest that while AWS may come with higher costs, its superior performance can justify the investment for organizations seeking high efficiency and reliability.

The findings of this study contribute to the theoretical framework surrounding cloud performance optimization and cloud provider selection, reinforcing the significance of performance metrics in assessing cloud infrastructure suitability. The comparative analysis of VM performance across AWS, Azure and GCP highlights the importance of computational efficiency, a central theme in cloud computing research. These results support theories positing that computational performance, particularly in CPU, memory and disk I/O, plays a critical role in the overall effectiveness of IaaS models. This research extends the theoretical understanding of how performance differences among major cloud providers impact organizational decision-making, particularly for applications with stringent processing requirements.

This study also adds to existing literature on benchmarking and evaluation methodologies in cloud environments. By using the Phoronix Test Suite, this research validates the tool’s utility in producing reliable, reproducible and objective measures of cloud VM performance. Such benchmarking frameworks are foundational in the theoretical discourse on performance assessment, as they provide standardized methodologies for comparing cloud infrastructure in diverse operational contexts. The findings reinforce the role of controlled benchmarking as a theoretical basis for developing performance-driven decision models in cloud computing.

Moreover, this research provides a theoretical lens on the scalability and reliability of cloud services, aligning with previous studies on cloud architecture’s adaptability to high-demand workloads. The superior performance observed with AWS suggests that theoretical models of scalability are crucial when evaluating cloud platforms for large-scale, real-time applications. By identifying performance as a deciding factor in cloud provider selection, this study enhances theoretical models that integrate performance, cost and security as key components in cloud adoption frameworks.

In conclusion, this research strengthens theoretical foundations in cloud performance evaluation, offering insights into the implications of performance-based cloud selection. This contribution is valuable for researchers aiming to refine predictive models and decision-making frameworks for cloud adoption, especially as cloud technology continues to evolve.

This study’s findings provide valuable insights for organizations and IT decision-makers selecting cloud-based VM solutions. The performance differences among AWS, Microsoft Azure and GCP indicate that organizations aiming to optimize their cloud infrastructure must carefully weigh their provider options. AWS consistently outperforms Azure and GCP in key areas such as CPU processing, memory handling, and disk I/O. This makes AWS appealing for organizations with high-performance needs or those running compute-intensive applications. However, specific use cases and existing infrastructure may still favor Azure or GCP for certain organizations.

Performance is crucial, but organizations must also consider cost implications. Despite AWS’s superior performance, it may come with higher costs. IT managers should evaluate their performance needs against each provider’s pricing models to balance computational power and cost efficiency. For organizations planning to scale, understanding each provider’s performance capabilities is essential. AWS’s advantage may offer better scalability for large workloads, whereas Azure and GCP might be more suitable for businesses requiring deep integration with specific software tools or those already embedded in the Microsoft or Google ecosystems.

VM performance directly impacts application performance and user experience. Organizations relying on real-time processing – such as those in financial services, health care or e-commerce – may see significant differences in application performance based on their chosen VM provider. AWS’s superior performance, particularly its lower latency and higher throughput, could enhance user experience and operational efficiency for time-sensitive applications.

Benchmarking tools, such as the Phoronix Test Suite used in this study, are crucial for organizations making cloud-related decisions. These tools allow businesses to assess cloud provider performance in a controlled, repeatable manner, ensuring their chosen platform meets specific performance requirements. By conducting tailored performance tests before committing to a provider, organizations can make more informed decisions, leading to better long-term outcomes.

Performance, while crucial, is not the only factor to consider when choosing a cloud provider. Organizations must also weigh security, compliance, availability and support. These nonperformance aspects are equally important in ensuring that the selected cloud infrastructure meets the organization’s broader operational needs. A holistic approach – one that evaluates both performance and nonperformance factors – will lead to more informed decisions when selecting cloud VMs.

Although this study provides valuable insights into the comparative performance of cloud VMs across AWS, Azure and GCP, there are limitations that should be considered in interpreting the findings. One primary limitation is the focus on performance metrics as the sole criteria for comparison. Although performance is critical for applications with high computational demands, other factors, such as cost, security features, compliance and regional availability, also play significant roles in cloud provider selection. Future research could expand the scope of comparison to include these additional criteria, offering a more holistic view of cloud service provider suitability.

In addition, this study relied on specific configurations of VMs, with Ubuntu 20.04 LTS selected as the reference OS. This choice, while ensuring a controlled and consistent testing environment, may not fully capture the performance variations that might arise with different OS s or VM configurations tailored to specific applications or workloads. Future studies could investigate the performance of alternative OS s or specialized VM configurations, providing insights into how different setups impact cloud VM performance across providers.

The benchmarking tool used, the Phoronix Test Suite, provided a robust framework for this analysis, but it is limited to a set of predefined tests, primarily focusing on CPU, memory, disk I/O and network performance. Additional tools or custom benchmarks could be considered in future research to capture other performance dimensions, such as latency under high network loads, failover capabilities and performance in distributed computing scenarios. Expanding the benchmark suite to include real-world application scenarios, such as database operations or machine learning tasks, could yield results that are more applicable to specific industry needs.

Finally, this study is limited by its static, point-in-time analysis. Cloud providers continuously optimize and upgrade their infrastructure, which can affect performance over time. A longitudinal study that periodically reassesses VM performance across providers would provide a more dynamic understanding of performance trends and help organizations make more informed decisions. Tracking these changes could reveal how cloud providers adapt to evolving computational demands, providing insights into the sustainability and scalability of their infrastructure.

The current study compared the VM performance in the IaaS models of Microsoft, Amazon and Google, the three largest cloud services providers. Linux Ubuntu 20.04 LTS was used as the reference OS distribution. Using the benchmarking Phoronix Test Suite, the study measured the performance of each solution. Thirteen individual tests from this suite were applied: 7zip, gzip, build-linux-kernel, himeno, t-test1, ramspeed, sockperf, perf-bench, blogbench, fio, sysbench, cryptsetup and network-loopkback. The tests examined VM components such as processor, memory and OS. The results showed that Amazon’s VM under the Amazon Web Service platform achieved the best results. This VM achieved an average score of 87.71%, measured according to the methodology presented in Section 2.3. The Microsoft and Google VMs produced similar scores to each other, although they were much lower than Amazon’s score. The VM running under Microsoft Azure achieved an average score of 77.62%; the VM running under GCP achieved 77.55%.

This study was conducted with support under the project IG409031 “Úloha interního auditu v kontextu COVID-19 a zvýšených hrozeb informační”, granted by the Prague University of Economics and Business, Faculty of Informatics and Statistics.